[BUG] Random hard lockups during idle on AMD Ryzen 5 5600X (amd_pstate)

[Тимур]

Hi Linus and the LKML community,

I am experiencing frequent random hard lockups on Arch Linux. The
freezes occur exclusively when the system is idle or under very light
usage (e.g., browsing text or sitting at the terminal).

Environment details:
OS: Arch Linux (x86_64)
Kernel version: 7.0.10-arch1-1
CPU: AMD Ryzen 5 5600X 6-Core Processor

I managed to capture a serial console dump right before the system
completely froze:

[   42.109382] mce: [Hardware Error]: Machine check events logged
[   42.109395] [Hardware Error]: Corrected error, no action required.
[   42.109401] [Hardware Error]: CPU:0 (19:21:0) Bank: 5: bea000000100010b
[   42.109410] [Hardware Error]: TSC: 0 ADDR: ffffffff81001300 MISC:
d01a000000000000
[   42.109418] [Hardware Error]: PROCESSOR 2:a20f10 AMD Ryzen 5 5600X
[   42.109425] BUG: unable to handle page fault for address: ffff888104ea7000
[   42.109432] #PF: supervisor read access in kernel mode
[   42.109438] #PF: error_code(0x0000) - not-present page
[   42.109444] PGD 0 P4D 0
[   42.109449] Oops: 0000 [#1] PREEMPT SMP NOPTI
[   42.109456] CPU: 3 UID: 0 PID: 0 Comm: swapper/3 Not tainted 7.0.10-arch1-1
[   42.109465] RIP: 0010:amd_pstate_update+0x72/0x140
[   42.109490] Call Trace:
[   42.109494]  <TASK>
[   42.109498]  ? __die+0x20/0x60
[   42.109505]  ? page_fault_oops+0x150/0x440
[   42.109513]  ? exc_page_fault+0x68/0x150
[   42.109521]  ? asm_exc_page_fault+0x22/0x30
[   42.109529]  ? amd_pstate_update+0x72/0x140
[   42.109537]  cpufreq_update_util+0x45/0x60
[   42.109545]  psi_task_change+0x82/0xb0
[   42.109553]  finish_task_switch.isra.0+0x8d/0x2a0
[   42.109562]  __schedule+0x30f/0x9a0
[   42.109571]  schedule+0x5b/0xa0
[   42.109578]  do_idle+0x1a2/0x220
[   42.109586]  cpu_startup_entry+0x25/0x30
[   42.109594]  start_secondary+0x11d/0x140
[   42.109603]  common_startup_64+0x13e/0x141
[   42.109612]  </TASK>

The issue seems to be related to C-states or the current amd_pstate
driver behavior on this specific microarchitecture. Disabling global
C-states in the UEFI prevents the crash, but it is suboptimal for
power saving.

Please let me know if you need me to test any debug patches or provide
more verbose syslogs.

Regards,
sinat